Research shows 45% of AI-generated code contains security flaws, with models struggling particularly with Cross-Site Scripting (failing 86% of the time) and generating insecure cryptographic implementations in 14% of cases.
Open the full topic